# Copyright 2023 Huawei Technologies Co., Ltd
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ============================================================================
"""adamw"""
from __future__ import absolute_import
from mindspore.ops import functional as F, composite as C, operations as P
from mindspore.common.parameter import Parameter
from mindspore.common.tensor import Tensor
import mindspore.common.dtype as mstype
from mindspore.experimental.optim.optimizer import Optimizer
from mindspore import ops
from mindspore import jit
_adamw_opt = C.MultitypeFuncGraph("adamw_opt")
op_mul = P.Mul()
op_pow = P.Pow()
op_sqrt = P.Sqrt()
op_maximum = P.Maximum()
hyper_map = C.HyperMap()
@jit
def prepare_func(lr, weight_decay, state_step, beta1, beta2):
weight_decay_new = 1 - lr * weight_decay
bias_correction1 = 1 - op_pow(beta1, state_step)
bias_correction2 = 1 - op_pow(beta2, state_step)
step_size = lr / bias_correction1
bias_correction2_sqrt = op_sqrt(bias_correction2)
return weight_decay_new, step_size, bias_correction2_sqrt
@_adamw_opt.register("Tensor", "Tensor", "Bool", "Float", "Tensor", "Float", "Float", "Tensor", "Tensor",
"Tensor", "Tensor", "Tensor")
def _run_adamw_opt(weight_decay_new, step_size, amsgrad, eps, bias_correction2_sqrt, beta1, beta2, param, grad,
exp_avg, exp_avg_sq, max_exp_avg_sq):
"""Apply adamw optimizer to the weight parameter."""
success = True
next_param = op_mul(param, weight_decay_new)
F.assign(exp_avg, op_mul(exp_avg, beta1) + op_mul(grad, 1 - beta1))
F.assign(exp_avg_sq, ops.addcmul(op_mul(exp_avg_sq, beta2), grad, grad, 1 - beta2))
if amsgrad:
next_max_exp_avg = op_maximum(max_exp_avg_sq, exp_avg_sq)
denom = op_sqrt(next_max_exp_avg) / bias_correction2_sqrt + eps
F.assign(max_exp_avg_sq, next_max_exp_avg)
else:
denom = op_sqrt(exp_avg_sq) / bias_correction2_sqrt + eps
return_param = next_param - op_mul(exp_avg / denom, step_size)
F.assign(param, return_param)
return success
[docs]class AdamW(Optimizer):
r"""
Implements Adam Weight Decay algorithm.
.. math::
\begin{aligned}
&\textbf{input} : \gamma \text{(lr)}, \: \beta_1, \beta_2
\text{(betas)}, \: \theta_0 \text{(params)}, \: f(\theta) \text{(objective)},
\: \epsilon \text{ (epsilon)} \\
&\hspace{13mm} \lambda \text{(weight decay)}, \: \textit{amsgrad},
\: \textit{maximize} \\
&\textbf{initialize} : m_0 \leftarrow 0 \text{ (first moment)}, v_0 \leftarrow 0
\text{ ( second moment)}, \: \widehat{v_0}^{max}\leftarrow 0 \\[-1.ex]
&\textbf{for} \: t=1 \: \textbf{to} \: \ldots \: \textbf{do} \\
&\hspace{5mm}\textbf{if} \: \textit{maximize}: \\
&\hspace{10mm}g_t \leftarrow -\nabla_{\theta} f_t (\theta_{t-1}) \\
&\hspace{5mm}\textbf{else} \\
&\hspace{10mm}g_t \leftarrow \nabla_{\theta} f_t (\theta_{t-1}) \\
&\hspace{5mm} \theta_t \leftarrow \theta_{t-1} - \gamma \lambda \theta_{t-1} \\
&\hspace{5mm}m_t \leftarrow \beta_1 m_{t-1} + (1 - \beta_1) g_t \\
&\hspace{5mm}v_t \leftarrow \beta_2 v_{t-1} + (1-\beta_2) g^2_t \\
&\hspace{5mm}\widehat{m_t} \leftarrow m_t/\big(1-\beta_1^t \big) \\
&\hspace{5mm}\widehat{v_t} \leftarrow v_t/\big(1-\beta_2^t \big) \\
&\hspace{5mm}\textbf{if} \: amsgrad \\
&\hspace{10mm}\widehat{v_t}^{max} \leftarrow \mathrm{max}(\widehat{v_t}^{max},
\widehat{v_t}) \\
&\hspace{10mm}\theta_t \leftarrow \theta_t - \gamma \widehat{m_t}/
\big(\sqrt{\widehat{v_t}^{max}} + \epsilon \big) \\
&\hspace{5mm}\textbf{else} \\
&\hspace{10mm}\theta_t \leftarrow \theta_t - \gamma \widehat{m_t}/
\big(\sqrt{\widehat{v_t}} + \epsilon \big) \\
&\bf{return} \: \theta_t \\[-1.ex]
\end{aligned}
.. warning::
This is an experimental optimizer API that is subject to change.
This module must be used with lr scheduler module in `LRScheduler Class
<https://www.mindspore.cn/docs/en/r2.3.0rc2/api_python/mindspore.experimental.html#lrscheduler-class>`_ .
Args:
params (Union[list(Parameter), list(dict)]): list of parameters to optimize or dicts defining
parameter groups
lr (Union[int, float, Tensor], optional): learning rate. Default: ``1e-3``.
betas (Tuple[float, float], optional): The exponential decay rate for the moment estimations.
Default: ``(0.9, 0.999)``.
eps (float, optional): term added to the denominator to improve
numerical stability. Default: ``1e-8``.
weight_decay (float, optional): weight decay (L2 penalty). Default: ``0.``.
amsgrad (bool, optional): whether to use the AMSGrad algorithm. Default: ``False``.
Keyword Args:
maximize (bool, optional): maximize the params based on the objective, instead of minimizing.
Default: ``False``.
Inputs:
- **gradients** (tuple[Tensor]) - The gradients of `params`.
Raises:
ValueError: If the learning rate is not int, float or Tensor.
ValueError: If the learning rate is less than 0.
ValueError: If the `eps` is less than 0.0.
ValueError: If the `betas` not in the range of [0, 1).
ValueError: If the `weight_decay` is less than 0.
Supported Platforms:
``Ascend`` ``GPU`` ``CPU``
Examples:
>>> import mindspore
>>> from mindspore import nn
>>> from mindspore.experimental import optim
>>> # Define the network structure of LeNet5. Refer to
>>> # https://gitee.com/mindspore/docs/blob/r2.3.0rc2/docs/mindspore/code/lenet.py
>>> net = LeNet5()
>>> loss_fn = nn.SoftmaxCrossEntropyWithLogits(sparse=True)
>>> optimizer = optim.AdamW(net.trainable_params(), lr=0.1)
>>> def forward_fn(data, label):
... logits = net(data)
... loss = loss_fn(logits, label)
... return loss, logits
>>> grad_fn = mindspore.value_and_grad(forward_fn, None, optimizer.parameters, has_aux=True)
>>> def train_step(data, label):
... (loss, _), grads = grad_fn(data, label)
... optimizer(grads)
... return loss
"""
def __init__(self, params, lr=1e-3, betas=(0.9, 0.999), eps=1e-8,
weight_decay=1e-2, amsgrad=False, *, maximize=False):
if lr < 0.0:
raise ValueError("Invalid learning rate: {}".format(lr))
if eps < 0.0:
raise ValueError("Invalid epsilon value: {}".format(eps))
if not 0.0 <= betas[0] < 1.0:
raise ValueError("Invalid beta parameter at index 0: {}".format(betas[0]))
if not 0.0 <= betas[1] < 1.0:
raise ValueError("Invalid beta parameter at index 1: {}".format(betas[1]))
if weight_decay < 0.0:
raise ValueError("Invalid weight_decay value: {}".format(weight_decay))
defaults = dict(lr=lr, betas=betas, eps=eps,
weight_decay=weight_decay, amsgrad=amsgrad,
maximize=maximize)
super(AdamW, self).__init__(params, defaults)
self.exp_avg = self.parameters.clone(prefix="exp_avg", init='zeros')
self.exp_avg_sq = self.parameters.clone(prefix="exp_avg_sq", init='zeros')
self.max_exp_avg_sq = self.parameters.clone(prefix="max_exp_avg_sq", init='zeros')
self.state_step = Parameter(Tensor(0, mstype.int32), "state_step")
self.increase_tensor = Tensor(1, mstype.int32)
self.assignadd = P.AssignAdd()
self.op_cast = P.Cast()
@jit
def implementation(self, lr, weight_decay, beta1, beta2, amsgrad, eps, grads, start_id, end_id):
"""Extract the common computing part for acceleration"""
weight_decay_new, step_size, bias_correction2_sqrt = prepare_func(lr, weight_decay,
self.state_step, beta1, beta2)
self.hyper_map(F.partial(_adamw_opt, weight_decay_new, step_size, amsgrad,
eps, bias_correction2_sqrt, beta1, beta2),
self.parameters[start_id: end_id], grads, self.exp_avg[start_id: end_id],
self.exp_avg_sq[start_id: end_id], self.max_exp_avg_sq[start_id: end_id])
return True
def construct(self, gradients):
self.assignadd(self.state_step, self.increase_tensor)
for group_id, group in enumerate(self.param_groups):
beta1, beta2 = group['betas']
start_id = self.group_start_id[group_id]
end_id = self.group_start_id[group_id + 1]
lr = self.lrs[group_id]
if isinstance(group.get("lr"), float):
lr = self.op_cast(group.get("lr"), mstype.float32)
grads = tuple([grad if not group.get("maximize") else F.neg(grad) for grad in gradients[start_id: end_id]])
self.implementation(lr, group.get("weight_decay"), beta1, beta2, group.get("amsgrad"), group.get("eps"),
grads, start_id, end_id)
return True